Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_mount.h"
13#include "xfs_ag.h"
14#include "xfs_inode.h"
15#include "xfs_errortag.h"
16#include "xfs_error.h"
17#include "xfs_icache.h"
18#include "xfs_trans.h"
19#include "xfs_ialloc.h"
20#include "xfs_dir2.h"
21#include "xfs_health.h"
22
23#include <linux/iversion.h>
24
25/*
26 * If we are doing readahead on an inode buffer, we might be in log recovery
27 * reading an inode allocation buffer that hasn't yet been replayed, and hence
28 * has not had the inode cores stamped into it. Hence for readahead, the buffer
29 * may be potentially invalid.
30 *
31 * If the readahead buffer is invalid, we need to mark it with an error and
32 * clear the DONE status of the buffer so that a followup read will re-read it
33 * from disk. We don't report the error otherwise to avoid warnings during log
34 * recovery and we don't get unnecessary panics on debug kernels. We use EIO here
35 * because all we want to do is say readahead failed; there is no-one to report
36 * the error to, so this will distinguish it from a non-ra verifier failure.
37 * Changes to this readahead error behaviour also need to be reflected in
38 * xfs_dquot_buf_readahead_verify().
39 */
40static void
41xfs_inode_buf_verify(
42 struct xfs_buf *bp,
43 bool readahead)
44{
45 struct xfs_mount *mp = bp->b_mount;
46 int i;
47 int ni;
48
49 /*
50 * Validate the magic number and version of every inode in the buffer
51 */
52 ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock;
53 for (i = 0; i < ni; i++) {
54 struct xfs_dinode *dip;
55 xfs_agino_t unlinked_ino;
56 int di_ok;
57
58 dip = xfs_buf_offset(bp, (i << mp->m_sb.sb_inodelog));
59 unlinked_ino = be32_to_cpu(dip->di_next_unlinked);
60 di_ok = xfs_verify_magic16(bp, dip->di_magic) &&
61 xfs_dinode_good_version(mp, dip->di_version) &&
62 xfs_verify_agino_or_null(bp->b_pag, unlinked_ino);
63 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
64 XFS_ERRTAG_ITOBP_INOTOBP))) {
65 if (readahead) {
66 bp->b_flags &= ~XBF_DONE;
67 xfs_buf_ioerror(bp, -EIO);
68 return;
69 }
70
71#ifdef DEBUG
72 xfs_alert(mp,
73 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
74 (unsigned long long)xfs_buf_daddr(bp), i,
75 be16_to_cpu(dip->di_magic));
76#endif
77 xfs_buf_verifier_error(bp, -EFSCORRUPTED,
78 __func__, dip, sizeof(*dip),
79 NULL);
80 return;
81 }
82 }
83}
84
85
86static void
87xfs_inode_buf_read_verify(
88 struct xfs_buf *bp)
89{
90 xfs_inode_buf_verify(bp, false);
91}
92
93static void
94xfs_inode_buf_readahead_verify(
95 struct xfs_buf *bp)
96{
97 xfs_inode_buf_verify(bp, true);
98}
99
100static void
101xfs_inode_buf_write_verify(
102 struct xfs_buf *bp)
103{
104 xfs_inode_buf_verify(bp, false);
105}
106
107const struct xfs_buf_ops xfs_inode_buf_ops = {
108 .name = "xfs_inode",
109 .magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
110 cpu_to_be16(XFS_DINODE_MAGIC) },
111 .verify_read = xfs_inode_buf_read_verify,
112 .verify_write = xfs_inode_buf_write_verify,
113};
114
115const struct xfs_buf_ops xfs_inode_buf_ra_ops = {
116 .name = "xfs_inode_ra",
117 .magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
118 cpu_to_be16(XFS_DINODE_MAGIC) },
119 .verify_read = xfs_inode_buf_readahead_verify,
120 .verify_write = xfs_inode_buf_write_verify,
121};
122
123
124/*
125 * This routine is called to map an inode to the buffer containing the on-disk
126 * version of the inode. It returns a pointer to the buffer containing the
127 * on-disk inode in the bpp parameter.
128 */
129int
130xfs_imap_to_bp(
131 struct xfs_mount *mp,
132 struct xfs_trans *tp,
133 struct xfs_imap *imap,
134 struct xfs_buf **bpp)
135{
136 int error;
137
138 error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
139 imap->im_len, XBF_UNMAPPED, bpp, &xfs_inode_buf_ops);
140 if (xfs_metadata_is_sick(error))
141 xfs_agno_mark_sick(mp, xfs_daddr_to_agno(mp, imap->im_blkno),
142 XFS_SICK_AG_INODES);
143 return error;
144}
145
146static inline struct timespec64 xfs_inode_decode_bigtime(uint64_t ts)
147{
148 struct timespec64 tv;
149 uint32_t n;
150
151 tv.tv_sec = xfs_bigtime_to_unix(div_u64_rem(ts, NSEC_PER_SEC, &n));
152 tv.tv_nsec = n;
153
154 return tv;
155}
156
157/* Convert an ondisk timestamp to an incore timestamp. */
158struct timespec64
159xfs_inode_from_disk_ts(
160 struct xfs_dinode *dip,
161 const xfs_timestamp_t ts)
162{
163 struct timespec64 tv;
164 struct xfs_legacy_timestamp *lts;
165
166 if (xfs_dinode_has_bigtime(dip))
167 return xfs_inode_decode_bigtime(be64_to_cpu(ts));
168
169 lts = (struct xfs_legacy_timestamp *)&ts;
170 tv.tv_sec = (int)be32_to_cpu(lts->t_sec);
171 tv.tv_nsec = (int)be32_to_cpu(lts->t_nsec);
172
173 return tv;
174}
175
176int
177xfs_inode_from_disk(
178 struct xfs_inode *ip,
179 struct xfs_dinode *from)
180{
181 struct inode *inode = VFS_I(ip);
182 int error;
183 xfs_failaddr_t fa;
184
185 ASSERT(ip->i_cowfp == NULL);
186
187 fa = xfs_dinode_verify(ip->i_mount, ip->i_ino, from);
188 if (fa) {
189 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "dinode", from,
190 sizeof(*from), fa);
191 return -EFSCORRUPTED;
192 }
193
194 /*
195 * First get the permanent information that is needed to allocate an
196 * inode. If the inode is unused, mode is zero and we shouldn't mess
197 * with the uninitialized part of it.
198 */
199 if (!xfs_has_v3inodes(ip->i_mount))
200 ip->i_flushiter = be16_to_cpu(from->di_flushiter);
201 inode->i_generation = be32_to_cpu(from->di_gen);
202 inode->i_mode = be16_to_cpu(from->di_mode);
203 if (!inode->i_mode)
204 return 0;
205
206 /*
207 * Convert v1 inodes immediately to v2 inode format as this is the
208 * minimum inode version format we support in the rest of the code.
209 * They will also be unconditionally written back to disk as v2 inodes.
210 */
211 if (unlikely(from->di_version == 1)) {
212 set_nlink(inode, be16_to_cpu(from->di_onlink));
213 ip->i_projid = 0;
214 } else {
215 set_nlink(inode, be32_to_cpu(from->di_nlink));
216 ip->i_projid = (prid_t)be16_to_cpu(from->di_projid_hi) << 16 |
217 be16_to_cpu(from->di_projid_lo);
218 }
219
220 i_uid_write(inode, be32_to_cpu(from->di_uid));
221 i_gid_write(inode, be32_to_cpu(from->di_gid));
222
223 /*
224 * Time is signed, so need to convert to signed 32 bit before
225 * storing in inode timestamp which may be 64 bit. Otherwise
226 * a time before epoch is converted to a time long after epoch
227 * on 64 bit systems.
228 */
229 inode_set_atime_to_ts(inode,
230 xfs_inode_from_disk_ts(from, from->di_atime));
231 inode_set_mtime_to_ts(inode,
232 xfs_inode_from_disk_ts(from, from->di_mtime));
233 inode_set_ctime_to_ts(inode,
234 xfs_inode_from_disk_ts(from, from->di_ctime));
235
236 ip->i_disk_size = be64_to_cpu(from->di_size);
237 ip->i_nblocks = be64_to_cpu(from->di_nblocks);
238 ip->i_extsize = be32_to_cpu(from->di_extsize);
239 ip->i_forkoff = from->di_forkoff;
240 ip->i_diflags = be16_to_cpu(from->di_flags);
241 ip->i_next_unlinked = be32_to_cpu(from->di_next_unlinked);
242
243 if (from->di_dmevmask || from->di_dmstate)
244 xfs_iflags_set(ip, XFS_IPRESERVE_DM_FIELDS);
245
246 if (xfs_has_v3inodes(ip->i_mount)) {
247 inode_set_iversion_queried(inode,
248 be64_to_cpu(from->di_changecount));
249 ip->i_crtime = xfs_inode_from_disk_ts(from, from->di_crtime);
250 ip->i_diflags2 = be64_to_cpu(from->di_flags2);
251 ip->i_cowextsize = be32_to_cpu(from->di_cowextsize);
252 }
253
254 error = xfs_iformat_data_fork(ip, from);
255 if (error)
256 return error;
257 if (from->di_forkoff) {
258 error = xfs_iformat_attr_fork(ip, from);
259 if (error)
260 goto out_destroy_data_fork;
261 }
262 if (xfs_is_reflink_inode(ip))
263 xfs_ifork_init_cow(ip);
264 return 0;
265
266out_destroy_data_fork:
267 xfs_idestroy_fork(&ip->i_df);
268 return error;
269}
270
271/* Convert an incore timestamp to an ondisk timestamp. */
272static inline xfs_timestamp_t
273xfs_inode_to_disk_ts(
274 struct xfs_inode *ip,
275 const struct timespec64 tv)
276{
277 struct xfs_legacy_timestamp *lts;
278 xfs_timestamp_t ts;
279
280 if (xfs_inode_has_bigtime(ip))
281 return cpu_to_be64(xfs_inode_encode_bigtime(tv));
282
283 lts = (struct xfs_legacy_timestamp *)&ts;
284 lts->t_sec = cpu_to_be32(tv.tv_sec);
285 lts->t_nsec = cpu_to_be32(tv.tv_nsec);
286
287 return ts;
288}
289
290static inline void
291xfs_inode_to_disk_iext_counters(
292 struct xfs_inode *ip,
293 struct xfs_dinode *to)
294{
295 if (xfs_inode_has_large_extent_counts(ip)) {
296 to->di_big_nextents = cpu_to_be64(xfs_ifork_nextents(&ip->i_df));
297 to->di_big_anextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_af));
298 /*
299 * We might be upgrading the inode to use larger extent counters
300 * than was previously used. Hence zero the unused field.
301 */
302 to->di_nrext64_pad = cpu_to_be16(0);
303 } else {
304 to->di_nextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_df));
305 to->di_anextents = cpu_to_be16(xfs_ifork_nextents(&ip->i_af));
306 }
307}
308
309void
310xfs_inode_to_disk(
311 struct xfs_inode *ip,
312 struct xfs_dinode *to,
313 xfs_lsn_t lsn)
314{
315 struct inode *inode = VFS_I(ip);
316
317 to->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
318 to->di_onlink = 0;
319
320 to->di_format = xfs_ifork_format(&ip->i_df);
321 to->di_uid = cpu_to_be32(i_uid_read(inode));
322 to->di_gid = cpu_to_be32(i_gid_read(inode));
323 to->di_projid_lo = cpu_to_be16(ip->i_projid & 0xffff);
324 to->di_projid_hi = cpu_to_be16(ip->i_projid >> 16);
325
326 to->di_atime = xfs_inode_to_disk_ts(ip, inode_get_atime(inode));
327 to->di_mtime = xfs_inode_to_disk_ts(ip, inode_get_mtime(inode));
328 to->di_ctime = xfs_inode_to_disk_ts(ip, inode_get_ctime(inode));
329 to->di_nlink = cpu_to_be32(inode->i_nlink);
330 to->di_gen = cpu_to_be32(inode->i_generation);
331 to->di_mode = cpu_to_be16(inode->i_mode);
332
333 to->di_size = cpu_to_be64(ip->i_disk_size);
334 to->di_nblocks = cpu_to_be64(ip->i_nblocks);
335 to->di_extsize = cpu_to_be32(ip->i_extsize);
336 to->di_forkoff = ip->i_forkoff;
337 to->di_aformat = xfs_ifork_format(&ip->i_af);
338 to->di_flags = cpu_to_be16(ip->i_diflags);
339
340 if (xfs_has_v3inodes(ip->i_mount)) {
341 to->di_version = 3;
342 to->di_changecount = cpu_to_be64(inode_peek_iversion(inode));
343 to->di_crtime = xfs_inode_to_disk_ts(ip, ip->i_crtime);
344 to->di_flags2 = cpu_to_be64(ip->i_diflags2);
345 to->di_cowextsize = cpu_to_be32(ip->i_cowextsize);
346 to->di_ino = cpu_to_be64(ip->i_ino);
347 to->di_lsn = cpu_to_be64(lsn);
348 memset(to->di_pad2, 0, sizeof(to->di_pad2));
349 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
350 to->di_v3_pad = 0;
351 } else {
352 to->di_version = 2;
353 to->di_flushiter = cpu_to_be16(ip->i_flushiter);
354 memset(to->di_v2_pad, 0, sizeof(to->di_v2_pad));
355 }
356
357 xfs_inode_to_disk_iext_counters(ip, to);
358}
359
360static xfs_failaddr_t
361xfs_dinode_verify_fork(
362 struct xfs_dinode *dip,
363 struct xfs_mount *mp,
364 int whichfork)
365{
366 xfs_extnum_t di_nextents;
367 xfs_extnum_t max_extents;
368 mode_t mode = be16_to_cpu(dip->di_mode);
369 uint32_t fork_size = XFS_DFORK_SIZE(dip, mp, whichfork);
370 uint32_t fork_format = XFS_DFORK_FORMAT(dip, whichfork);
371
372 di_nextents = xfs_dfork_nextents(dip, whichfork);
373
374 /*
375 * For fork types that can contain local data, check that the fork
376 * format matches the size of local data contained within the fork.
377 *
378 * For all types, check that when the size says the should be in extent
379 * or btree format, the inode isn't claiming it is in local format.
380 */
381 if (whichfork == XFS_DATA_FORK) {
382 if (S_ISDIR(mode) || S_ISLNK(mode)) {
383 if (be64_to_cpu(dip->di_size) <= fork_size &&
384 fork_format != XFS_DINODE_FMT_LOCAL)
385 return __this_address;
386 }
387
388 if (be64_to_cpu(dip->di_size) > fork_size &&
389 fork_format == XFS_DINODE_FMT_LOCAL)
390 return __this_address;
391 }
392
393 switch (fork_format) {
394 case XFS_DINODE_FMT_LOCAL:
395 /*
396 * No local regular files yet.
397 */
398 if (S_ISREG(mode) && whichfork == XFS_DATA_FORK)
399 return __this_address;
400 if (di_nextents)
401 return __this_address;
402 break;
403 case XFS_DINODE_FMT_EXTENTS:
404 if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
405 return __this_address;
406 break;
407 case XFS_DINODE_FMT_BTREE:
408 max_extents = xfs_iext_max_nextents(
409 xfs_dinode_has_large_extent_counts(dip),
410 whichfork);
411 if (di_nextents > max_extents)
412 return __this_address;
413 break;
414 default:
415 return __this_address;
416 }
417 return NULL;
418}
419
420static xfs_failaddr_t
421xfs_dinode_verify_forkoff(
422 struct xfs_dinode *dip,
423 struct xfs_mount *mp)
424{
425 if (!dip->di_forkoff)
426 return NULL;
427
428 switch (dip->di_format) {
429 case XFS_DINODE_FMT_DEV:
430 if (dip->di_forkoff != (roundup(sizeof(xfs_dev_t), 8) >> 3))
431 return __this_address;
432 break;
433 case XFS_DINODE_FMT_LOCAL: /* fall through ... */
434 case XFS_DINODE_FMT_EXTENTS: /* fall through ... */
435 case XFS_DINODE_FMT_BTREE:
436 if (dip->di_forkoff >= (XFS_LITINO(mp) >> 3))
437 return __this_address;
438 break;
439 default:
440 return __this_address;
441 }
442 return NULL;
443}
444
445static xfs_failaddr_t
446xfs_dinode_verify_nrext64(
447 struct xfs_mount *mp,
448 struct xfs_dinode *dip)
449{
450 if (xfs_dinode_has_large_extent_counts(dip)) {
451 if (!xfs_has_large_extent_counts(mp))
452 return __this_address;
453 if (dip->di_nrext64_pad != 0)
454 return __this_address;
455 } else if (dip->di_version >= 3) {
456 if (dip->di_v3_pad != 0)
457 return __this_address;
458 }
459
460 return NULL;
461}
462
463xfs_failaddr_t
464xfs_dinode_verify(
465 struct xfs_mount *mp,
466 xfs_ino_t ino,
467 struct xfs_dinode *dip)
468{
469 xfs_failaddr_t fa;
470 uint16_t mode;
471 uint16_t flags;
472 uint64_t flags2;
473 uint64_t di_size;
474 xfs_extnum_t nextents;
475 xfs_extnum_t naextents;
476 xfs_filblks_t nblocks;
477
478 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
479 return __this_address;
480
481 /* Verify v3 integrity information first */
482 if (dip->di_version >= 3) {
483 if (!xfs_has_v3inodes(mp))
484 return __this_address;
485 if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
486 XFS_DINODE_CRC_OFF))
487 return __this_address;
488 if (be64_to_cpu(dip->di_ino) != ino)
489 return __this_address;
490 if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
491 return __this_address;
492 }
493
494 /* don't allow invalid i_size */
495 di_size = be64_to_cpu(dip->di_size);
496 if (di_size & (1ULL << 63))
497 return __this_address;
498
499 mode = be16_to_cpu(dip->di_mode);
500 if (mode && xfs_mode_to_ftype(mode) == XFS_DIR3_FT_UNKNOWN)
501 return __this_address;
502
503 /* No zero-length symlinks/dirs. */
504 if ((S_ISLNK(mode) || S_ISDIR(mode)) && di_size == 0)
505 return __this_address;
506
507 fa = xfs_dinode_verify_nrext64(mp, dip);
508 if (fa)
509 return fa;
510
511 nextents = xfs_dfork_data_extents(dip);
512 naextents = xfs_dfork_attr_extents(dip);
513 nblocks = be64_to_cpu(dip->di_nblocks);
514
515 /* Fork checks carried over from xfs_iformat_fork */
516 if (mode && nextents + naextents > nblocks)
517 return __this_address;
518
519 if (nextents + naextents == 0 && nblocks != 0)
520 return __this_address;
521
522 if (S_ISDIR(mode) && nextents > mp->m_dir_geo->max_extents)
523 return __this_address;
524
525 if (mode && XFS_DFORK_BOFF(dip) > mp->m_sb.sb_inodesize)
526 return __this_address;
527
528 flags = be16_to_cpu(dip->di_flags);
529
530 if (mode && (flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
531 return __this_address;
532
533 /* check for illegal values of forkoff */
534 fa = xfs_dinode_verify_forkoff(dip, mp);
535 if (fa)
536 return fa;
537
538 /* Do we have appropriate data fork formats for the mode? */
539 switch (mode & S_IFMT) {
540 case S_IFIFO:
541 case S_IFCHR:
542 case S_IFBLK:
543 case S_IFSOCK:
544 if (dip->di_format != XFS_DINODE_FMT_DEV)
545 return __this_address;
546 break;
547 case S_IFREG:
548 case S_IFLNK:
549 case S_IFDIR:
550 fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
551 if (fa)
552 return fa;
553 break;
554 case 0:
555 /* Uninitialized inode ok. */
556 break;
557 default:
558 return __this_address;
559 }
560
561 if (dip->di_forkoff) {
562 fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
563 if (fa)
564 return fa;
565 } else {
566 /*
567 * If there is no fork offset, this may be a freshly-made inode
568 * in a new disk cluster, in which case di_aformat is zeroed.
569 * Otherwise, such an inode must be in EXTENTS format; this goes
570 * for freed inodes as well.
571 */
572 switch (dip->di_aformat) {
573 case 0:
574 case XFS_DINODE_FMT_EXTENTS:
575 break;
576 default:
577 return __this_address;
578 }
579 if (naextents)
580 return __this_address;
581 }
582
583 /* extent size hint validation */
584 fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),
585 mode, flags);
586 if (fa)
587 return fa;
588
589 /* only version 3 or greater inodes are extensively verified here */
590 if (dip->di_version < 3)
591 return NULL;
592
593 flags2 = be64_to_cpu(dip->di_flags2);
594
595 /* don't allow reflink/cowextsize if we don't have reflink */
596 if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
597 !xfs_has_reflink(mp))
598 return __this_address;
599
600 /* only regular files get reflink */
601 if ((flags2 & XFS_DIFLAG2_REFLINK) && (mode & S_IFMT) != S_IFREG)
602 return __this_address;
603
604 /* don't let reflink and realtime mix */
605 if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME))
606 return __this_address;
607
608 /* COW extent size hint validation */
609 fa = xfs_inode_validate_cowextsize(mp, be32_to_cpu(dip->di_cowextsize),
610 mode, flags, flags2);
611 if (fa)
612 return fa;
613
614 /* bigtime iflag can only happen on bigtime filesystems */
615 if (xfs_dinode_has_bigtime(dip) &&
616 !xfs_has_bigtime(mp))
617 return __this_address;
618
619 return NULL;
620}
621
622void
623xfs_dinode_calc_crc(
624 struct xfs_mount *mp,
625 struct xfs_dinode *dip)
626{
627 uint32_t crc;
628
629 if (dip->di_version < 3)
630 return;
631
632 ASSERT(xfs_has_crc(mp));
633 crc = xfs_start_cksum_update((char *)dip, mp->m_sb.sb_inodesize,
634 XFS_DINODE_CRC_OFF);
635 dip->di_crc = xfs_end_cksum(crc);
636}
637
638/*
639 * Validate di_extsize hint.
640 *
641 * 1. Extent size hint is only valid for directories and regular files.
642 * 2. FS_XFLAG_EXTSIZE is only valid for regular files.
643 * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
644 * 4. Hint cannot be larger than MAXTEXTLEN.
645 * 5. Can be changed on directories at any time.
646 * 6. Hint value of 0 turns off hints, clears inode flags.
647 * 7. Extent size must be a multiple of the appropriate block size.
648 * For realtime files, this is the rt extent size.
649 * 8. For non-realtime files, the extent size hint must be limited
650 * to half the AG size to avoid alignment extending the extent beyond the
651 * limits of the AG.
652 */
653xfs_failaddr_t
654xfs_inode_validate_extsize(
655 struct xfs_mount *mp,
656 uint32_t extsize,
657 uint16_t mode,
658 uint16_t flags)
659{
660 bool rt_flag;
661 bool hint_flag;
662 bool inherit_flag;
663 uint32_t extsize_bytes;
664 uint32_t blocksize_bytes;
665
666 rt_flag = (flags & XFS_DIFLAG_REALTIME);
667 hint_flag = (flags & XFS_DIFLAG_EXTSIZE);
668 inherit_flag = (flags & XFS_DIFLAG_EXTSZINHERIT);
669 extsize_bytes = XFS_FSB_TO_B(mp, extsize);
670
671 /*
672 * This comment describes a historic gap in this verifier function.
673 *
674 * For a directory with both RTINHERIT and EXTSZINHERIT flags set, this
675 * function has never checked that the extent size hint is an integer
676 * multiple of the realtime extent size. Since we allow users to set
677 * this combination on non-rt filesystems /and/ to change the rt
678 * extent size when adding a rt device to a filesystem, the net effect
679 * is that users can configure a filesystem anticipating one rt
680 * geometry and change their minds later. Directories do not use the
681 * extent size hint, so this is harmless for them.
682 *
683 * If a directory with a misaligned extent size hint is allowed to
684 * propagate that hint into a new regular realtime file, the result
685 * is that the inode cluster buffer verifier will trigger a corruption
686 * shutdown the next time it is run, because the verifier has always
687 * enforced the alignment rule for regular files.
688 *
689 * Because we allow administrators to set a new rt extent size when
690 * adding a rt section, we cannot add a check to this verifier because
691 * that will result a new source of directory corruption errors when
692 * reading an existing filesystem. Instead, we rely on callers to
693 * decide when alignment checks are appropriate, and fix things up as
694 * needed.
695 */
696
697 if (rt_flag)
698 blocksize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
699 else
700 blocksize_bytes = mp->m_sb.sb_blocksize;
701
702 if ((hint_flag || inherit_flag) && !(S_ISDIR(mode) || S_ISREG(mode)))
703 return __this_address;
704
705 if (hint_flag && !S_ISREG(mode))
706 return __this_address;
707
708 if (inherit_flag && !S_ISDIR(mode))
709 return __this_address;
710
711 if ((hint_flag || inherit_flag) && extsize == 0)
712 return __this_address;
713
714 /* free inodes get flags set to zero but extsize remains */
715 if (mode && !(hint_flag || inherit_flag) && extsize != 0)
716 return __this_address;
717
718 if (extsize_bytes % blocksize_bytes)
719 return __this_address;
720
721 if (extsize > XFS_MAX_BMBT_EXTLEN)
722 return __this_address;
723
724 if (!rt_flag && extsize > mp->m_sb.sb_agblocks / 2)
725 return __this_address;
726
727 return NULL;
728}
729
730/*
731 * Validate di_cowextsize hint.
732 *
733 * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
734 * The inode does not have to have any shared blocks, but it must be a v3.
735 * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
736 * for a directory, the hint is propagated to new files.
737 * 3. Can be changed on files & directories at any time.
738 * 4. Hint value of 0 turns off hints, clears inode flags.
739 * 5. Extent size must be a multiple of the appropriate block size.
740 * 6. The extent size hint must be limited to half the AG size to avoid
741 * alignment extending the extent beyond the limits of the AG.
742 */
743xfs_failaddr_t
744xfs_inode_validate_cowextsize(
745 struct xfs_mount *mp,
746 uint32_t cowextsize,
747 uint16_t mode,
748 uint16_t flags,
749 uint64_t flags2)
750{
751 bool rt_flag;
752 bool hint_flag;
753 uint32_t cowextsize_bytes;
754
755 rt_flag = (flags & XFS_DIFLAG_REALTIME);
756 hint_flag = (flags2 & XFS_DIFLAG2_COWEXTSIZE);
757 cowextsize_bytes = XFS_FSB_TO_B(mp, cowextsize);
758
759 if (hint_flag && !xfs_has_reflink(mp))
760 return __this_address;
761
762 if (hint_flag && !(S_ISDIR(mode) || S_ISREG(mode)))
763 return __this_address;
764
765 if (hint_flag && cowextsize == 0)
766 return __this_address;
767
768 /* free inodes get flags set to zero but cowextsize remains */
769 if (mode && !hint_flag && cowextsize != 0)
770 return __this_address;
771
772 if (hint_flag && rt_flag)
773 return __this_address;
774
775 if (cowextsize_bytes % mp->m_sb.sb_blocksize)
776 return __this_address;
777
778 if (cowextsize > XFS_MAX_BMBT_EXTLEN)
779 return __this_address;
780
781 if (cowextsize > mp->m_sb.sb_agblocks / 2)
782 return __this_address;
783
784 return NULL;
785}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_mount.h"
13#include "xfs_inode.h"
14#include "xfs_errortag.h"
15#include "xfs_error.h"
16#include "xfs_icache.h"
17#include "xfs_trans.h"
18#include "xfs_ialloc.h"
19#include "xfs_dir2.h"
20
21#include <linux/iversion.h>
22
23/*
24 * If we are doing readahead on an inode buffer, we might be in log recovery
25 * reading an inode allocation buffer that hasn't yet been replayed, and hence
26 * has not had the inode cores stamped into it. Hence for readahead, the buffer
27 * may be potentially invalid.
28 *
29 * If the readahead buffer is invalid, we need to mark it with an error and
30 * clear the DONE status of the buffer so that a followup read will re-read it
31 * from disk. We don't report the error otherwise to avoid warnings during log
32 * recovery and we don't get unnecessary panics on debug kernels. We use EIO here
33 * because all we want to do is say readahead failed; there is no-one to report
34 * the error to, so this will distinguish it from a non-ra verifier failure.
35 * Changes to this readahead error behaviour also need to be reflected in
36 * xfs_dquot_buf_readahead_verify().
37 */
38static void
39xfs_inode_buf_verify(
40 struct xfs_buf *bp,
41 bool readahead)
42{
43 struct xfs_mount *mp = bp->b_mount;
44 xfs_agnumber_t agno;
45 int i;
46 int ni;
47
48 /*
49 * Validate the magic number and version of every inode in the buffer
50 */
51 agno = xfs_daddr_to_agno(mp, XFS_BUF_ADDR(bp));
52 ni = XFS_BB_TO_FSB(mp, bp->b_length) * mp->m_sb.sb_inopblock;
53 for (i = 0; i < ni; i++) {
54 int di_ok;
55 xfs_dinode_t *dip;
56 xfs_agino_t unlinked_ino;
57
58 dip = xfs_buf_offset(bp, (i << mp->m_sb.sb_inodelog));
59 unlinked_ino = be32_to_cpu(dip->di_next_unlinked);
60 di_ok = xfs_verify_magic16(bp, dip->di_magic) &&
61 xfs_dinode_good_version(&mp->m_sb, dip->di_version) &&
62 xfs_verify_agino_or_null(mp, agno, unlinked_ino);
63 if (unlikely(XFS_TEST_ERROR(!di_ok, mp,
64 XFS_ERRTAG_ITOBP_INOTOBP))) {
65 if (readahead) {
66 bp->b_flags &= ~XBF_DONE;
67 xfs_buf_ioerror(bp, -EIO);
68 return;
69 }
70
71#ifdef DEBUG
72 xfs_alert(mp,
73 "bad inode magic/vsn daddr %lld #%d (magic=%x)",
74 (unsigned long long)bp->b_bn, i,
75 be16_to_cpu(dip->di_magic));
76#endif
77 xfs_buf_verifier_error(bp, -EFSCORRUPTED,
78 __func__, dip, sizeof(*dip),
79 NULL);
80 return;
81 }
82 }
83}
84
85
86static void
87xfs_inode_buf_read_verify(
88 struct xfs_buf *bp)
89{
90 xfs_inode_buf_verify(bp, false);
91}
92
93static void
94xfs_inode_buf_readahead_verify(
95 struct xfs_buf *bp)
96{
97 xfs_inode_buf_verify(bp, true);
98}
99
100static void
101xfs_inode_buf_write_verify(
102 struct xfs_buf *bp)
103{
104 xfs_inode_buf_verify(bp, false);
105}
106
107const struct xfs_buf_ops xfs_inode_buf_ops = {
108 .name = "xfs_inode",
109 .magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
110 cpu_to_be16(XFS_DINODE_MAGIC) },
111 .verify_read = xfs_inode_buf_read_verify,
112 .verify_write = xfs_inode_buf_write_verify,
113};
114
115const struct xfs_buf_ops xfs_inode_buf_ra_ops = {
116 .name = "xfs_inode_ra",
117 .magic16 = { cpu_to_be16(XFS_DINODE_MAGIC),
118 cpu_to_be16(XFS_DINODE_MAGIC) },
119 .verify_read = xfs_inode_buf_readahead_verify,
120 .verify_write = xfs_inode_buf_write_verify,
121};
122
123
124/*
125 * This routine is called to map an inode to the buffer containing the on-disk
126 * version of the inode. It returns a pointer to the buffer containing the
127 * on-disk inode in the bpp parameter.
128 */
129int
130xfs_imap_to_bp(
131 struct xfs_mount *mp,
132 struct xfs_trans *tp,
133 struct xfs_imap *imap,
134 struct xfs_buf **bpp)
135{
136 return xfs_trans_read_buf(mp, tp, mp->m_ddev_targp, imap->im_blkno,
137 imap->im_len, XBF_UNMAPPED, bpp,
138 &xfs_inode_buf_ops);
139}
140
141static inline struct timespec64 xfs_inode_decode_bigtime(uint64_t ts)
142{
143 struct timespec64 tv;
144 uint32_t n;
145
146 tv.tv_sec = xfs_bigtime_to_unix(div_u64_rem(ts, NSEC_PER_SEC, &n));
147 tv.tv_nsec = n;
148
149 return tv;
150}
151
152/* Convert an ondisk timestamp to an incore timestamp. */
153struct timespec64
154xfs_inode_from_disk_ts(
155 struct xfs_dinode *dip,
156 const xfs_timestamp_t ts)
157{
158 struct timespec64 tv;
159 struct xfs_legacy_timestamp *lts;
160
161 if (xfs_dinode_has_bigtime(dip))
162 return xfs_inode_decode_bigtime(be64_to_cpu(ts));
163
164 lts = (struct xfs_legacy_timestamp *)&ts;
165 tv.tv_sec = (int)be32_to_cpu(lts->t_sec);
166 tv.tv_nsec = (int)be32_to_cpu(lts->t_nsec);
167
168 return tv;
169}
170
171int
172xfs_inode_from_disk(
173 struct xfs_inode *ip,
174 struct xfs_dinode *from)
175{
176 struct inode *inode = VFS_I(ip);
177 int error;
178 xfs_failaddr_t fa;
179
180 ASSERT(ip->i_cowfp == NULL);
181 ASSERT(ip->i_afp == NULL);
182
183 fa = xfs_dinode_verify(ip->i_mount, ip->i_ino, from);
184 if (fa) {
185 xfs_inode_verifier_error(ip, -EFSCORRUPTED, "dinode", from,
186 sizeof(*from), fa);
187 return -EFSCORRUPTED;
188 }
189
190 /*
191 * First get the permanent information that is needed to allocate an
192 * inode. If the inode is unused, mode is zero and we shouldn't mess
193 * with the uninitialized part of it.
194 */
195 if (!xfs_sb_version_has_v3inode(&ip->i_mount->m_sb))
196 ip->i_flushiter = be16_to_cpu(from->di_flushiter);
197 inode->i_generation = be32_to_cpu(from->di_gen);
198 inode->i_mode = be16_to_cpu(from->di_mode);
199 if (!inode->i_mode)
200 return 0;
201
202 /*
203 * Convert v1 inodes immediately to v2 inode format as this is the
204 * minimum inode version format we support in the rest of the code.
205 * They will also be unconditionally written back to disk as v2 inodes.
206 */
207 if (unlikely(from->di_version == 1)) {
208 set_nlink(inode, be16_to_cpu(from->di_onlink));
209 ip->i_projid = 0;
210 } else {
211 set_nlink(inode, be32_to_cpu(from->di_nlink));
212 ip->i_projid = (prid_t)be16_to_cpu(from->di_projid_hi) << 16 |
213 be16_to_cpu(from->di_projid_lo);
214 }
215
216 i_uid_write(inode, be32_to_cpu(from->di_uid));
217 i_gid_write(inode, be32_to_cpu(from->di_gid));
218
219 /*
220 * Time is signed, so need to convert to signed 32 bit before
221 * storing in inode timestamp which may be 64 bit. Otherwise
222 * a time before epoch is converted to a time long after epoch
223 * on 64 bit systems.
224 */
225 inode->i_atime = xfs_inode_from_disk_ts(from, from->di_atime);
226 inode->i_mtime = xfs_inode_from_disk_ts(from, from->di_mtime);
227 inode->i_ctime = xfs_inode_from_disk_ts(from, from->di_ctime);
228
229 ip->i_disk_size = be64_to_cpu(from->di_size);
230 ip->i_nblocks = be64_to_cpu(from->di_nblocks);
231 ip->i_extsize = be32_to_cpu(from->di_extsize);
232 ip->i_forkoff = from->di_forkoff;
233 ip->i_diflags = be16_to_cpu(from->di_flags);
234
235 if (from->di_dmevmask || from->di_dmstate)
236 xfs_iflags_set(ip, XFS_IPRESERVE_DM_FIELDS);
237
238 if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
239 inode_set_iversion_queried(inode,
240 be64_to_cpu(from->di_changecount));
241 ip->i_crtime = xfs_inode_from_disk_ts(from, from->di_crtime);
242 ip->i_diflags2 = be64_to_cpu(from->di_flags2);
243 ip->i_cowextsize = be32_to_cpu(from->di_cowextsize);
244 }
245
246 error = xfs_iformat_data_fork(ip, from);
247 if (error)
248 return error;
249 if (from->di_forkoff) {
250 error = xfs_iformat_attr_fork(ip, from);
251 if (error)
252 goto out_destroy_data_fork;
253 }
254 if (xfs_is_reflink_inode(ip))
255 xfs_ifork_init_cow(ip);
256 return 0;
257
258out_destroy_data_fork:
259 xfs_idestroy_fork(&ip->i_df);
260 return error;
261}
262
263/* Convert an incore timestamp to an ondisk timestamp. */
264static inline xfs_timestamp_t
265xfs_inode_to_disk_ts(
266 struct xfs_inode *ip,
267 const struct timespec64 tv)
268{
269 struct xfs_legacy_timestamp *lts;
270 xfs_timestamp_t ts;
271
272 if (xfs_inode_has_bigtime(ip))
273 return cpu_to_be64(xfs_inode_encode_bigtime(tv));
274
275 lts = (struct xfs_legacy_timestamp *)&ts;
276 lts->t_sec = cpu_to_be32(tv.tv_sec);
277 lts->t_nsec = cpu_to_be32(tv.tv_nsec);
278
279 return ts;
280}
281
282void
283xfs_inode_to_disk(
284 struct xfs_inode *ip,
285 struct xfs_dinode *to,
286 xfs_lsn_t lsn)
287{
288 struct inode *inode = VFS_I(ip);
289
290 to->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
291 to->di_onlink = 0;
292
293 to->di_format = xfs_ifork_format(&ip->i_df);
294 to->di_uid = cpu_to_be32(i_uid_read(inode));
295 to->di_gid = cpu_to_be32(i_gid_read(inode));
296 to->di_projid_lo = cpu_to_be16(ip->i_projid & 0xffff);
297 to->di_projid_hi = cpu_to_be16(ip->i_projid >> 16);
298
299 memset(to->di_pad, 0, sizeof(to->di_pad));
300 to->di_atime = xfs_inode_to_disk_ts(ip, inode->i_atime);
301 to->di_mtime = xfs_inode_to_disk_ts(ip, inode->i_mtime);
302 to->di_ctime = xfs_inode_to_disk_ts(ip, inode->i_ctime);
303 to->di_nlink = cpu_to_be32(inode->i_nlink);
304 to->di_gen = cpu_to_be32(inode->i_generation);
305 to->di_mode = cpu_to_be16(inode->i_mode);
306
307 to->di_size = cpu_to_be64(ip->i_disk_size);
308 to->di_nblocks = cpu_to_be64(ip->i_nblocks);
309 to->di_extsize = cpu_to_be32(ip->i_extsize);
310 to->di_nextents = cpu_to_be32(xfs_ifork_nextents(&ip->i_df));
311 to->di_anextents = cpu_to_be16(xfs_ifork_nextents(ip->i_afp));
312 to->di_forkoff = ip->i_forkoff;
313 to->di_aformat = xfs_ifork_format(ip->i_afp);
314 to->di_flags = cpu_to_be16(ip->i_diflags);
315
316 if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
317 to->di_version = 3;
318 to->di_changecount = cpu_to_be64(inode_peek_iversion(inode));
319 to->di_crtime = xfs_inode_to_disk_ts(ip, ip->i_crtime);
320 to->di_flags2 = cpu_to_be64(ip->i_diflags2);
321 to->di_cowextsize = cpu_to_be32(ip->i_cowextsize);
322 to->di_ino = cpu_to_be64(ip->i_ino);
323 to->di_lsn = cpu_to_be64(lsn);
324 memset(to->di_pad2, 0, sizeof(to->di_pad2));
325 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);
326 to->di_flushiter = 0;
327 } else {
328 to->di_version = 2;
329 to->di_flushiter = cpu_to_be16(ip->i_flushiter);
330 }
331}
332
333static xfs_failaddr_t
334xfs_dinode_verify_fork(
335 struct xfs_dinode *dip,
336 struct xfs_mount *mp,
337 int whichfork)
338{
339 uint32_t di_nextents = XFS_DFORK_NEXTENTS(dip, whichfork);
340
341 switch (XFS_DFORK_FORMAT(dip, whichfork)) {
342 case XFS_DINODE_FMT_LOCAL:
343 /*
344 * no local regular files yet
345 */
346 if (whichfork == XFS_DATA_FORK) {
347 if (S_ISREG(be16_to_cpu(dip->di_mode)))
348 return __this_address;
349 if (be64_to_cpu(dip->di_size) >
350 XFS_DFORK_SIZE(dip, mp, whichfork))
351 return __this_address;
352 }
353 if (di_nextents)
354 return __this_address;
355 break;
356 case XFS_DINODE_FMT_EXTENTS:
357 if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
358 return __this_address;
359 break;
360 case XFS_DINODE_FMT_BTREE:
361 if (whichfork == XFS_ATTR_FORK) {
362 if (di_nextents > MAXAEXTNUM)
363 return __this_address;
364 } else if (di_nextents > MAXEXTNUM) {
365 return __this_address;
366 }
367 break;
368 default:
369 return __this_address;
370 }
371 return NULL;
372}
373
374static xfs_failaddr_t
375xfs_dinode_verify_forkoff(
376 struct xfs_dinode *dip,
377 struct xfs_mount *mp)
378{
379 if (!dip->di_forkoff)
380 return NULL;
381
382 switch (dip->di_format) {
383 case XFS_DINODE_FMT_DEV:
384 if (dip->di_forkoff != (roundup(sizeof(xfs_dev_t), 8) >> 3))
385 return __this_address;
386 break;
387 case XFS_DINODE_FMT_LOCAL: /* fall through ... */
388 case XFS_DINODE_FMT_EXTENTS: /* fall through ... */
389 case XFS_DINODE_FMT_BTREE:
390 if (dip->di_forkoff >= (XFS_LITINO(mp) >> 3))
391 return __this_address;
392 break;
393 default:
394 return __this_address;
395 }
396 return NULL;
397}
398
399xfs_failaddr_t
400xfs_dinode_verify(
401 struct xfs_mount *mp,
402 xfs_ino_t ino,
403 struct xfs_dinode *dip)
404{
405 xfs_failaddr_t fa;
406 uint16_t mode;
407 uint16_t flags;
408 uint64_t flags2;
409 uint64_t di_size;
410
411 if (dip->di_magic != cpu_to_be16(XFS_DINODE_MAGIC))
412 return __this_address;
413
414 /* Verify v3 integrity information first */
415 if (dip->di_version >= 3) {
416 if (!xfs_sb_version_has_v3inode(&mp->m_sb))
417 return __this_address;
418 if (!xfs_verify_cksum((char *)dip, mp->m_sb.sb_inodesize,
419 XFS_DINODE_CRC_OFF))
420 return __this_address;
421 if (be64_to_cpu(dip->di_ino) != ino)
422 return __this_address;
423 if (!uuid_equal(&dip->di_uuid, &mp->m_sb.sb_meta_uuid))
424 return __this_address;
425 }
426
427 /* don't allow invalid i_size */
428 di_size = be64_to_cpu(dip->di_size);
429 if (di_size & (1ULL << 63))
430 return __this_address;
431
432 mode = be16_to_cpu(dip->di_mode);
433 if (mode && xfs_mode_to_ftype(mode) == XFS_DIR3_FT_UNKNOWN)
434 return __this_address;
435
436 /* No zero-length symlinks/dirs. */
437 if ((S_ISLNK(mode) || S_ISDIR(mode)) && di_size == 0)
438 return __this_address;
439
440 /* Fork checks carried over from xfs_iformat_fork */
441 if (mode &&
442 be32_to_cpu(dip->di_nextents) + be16_to_cpu(dip->di_anextents) >
443 be64_to_cpu(dip->di_nblocks))
444 return __this_address;
445
446 if (mode && XFS_DFORK_BOFF(dip) > mp->m_sb.sb_inodesize)
447 return __this_address;
448
449 flags = be16_to_cpu(dip->di_flags);
450
451 if (mode && (flags & XFS_DIFLAG_REALTIME) && !mp->m_rtdev_targp)
452 return __this_address;
453
454 /* check for illegal values of forkoff */
455 fa = xfs_dinode_verify_forkoff(dip, mp);
456 if (fa)
457 return fa;
458
459 /* Do we have appropriate data fork formats for the mode? */
460 switch (mode & S_IFMT) {
461 case S_IFIFO:
462 case S_IFCHR:
463 case S_IFBLK:
464 case S_IFSOCK:
465 if (dip->di_format != XFS_DINODE_FMT_DEV)
466 return __this_address;
467 break;
468 case S_IFREG:
469 case S_IFLNK:
470 case S_IFDIR:
471 fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
472 if (fa)
473 return fa;
474 break;
475 case 0:
476 /* Uninitialized inode ok. */
477 break;
478 default:
479 return __this_address;
480 }
481
482 if (dip->di_forkoff) {
483 fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
484 if (fa)
485 return fa;
486 } else {
487 /*
488 * If there is no fork offset, this may be a freshly-made inode
489 * in a new disk cluster, in which case di_aformat is zeroed.
490 * Otherwise, such an inode must be in EXTENTS format; this goes
491 * for freed inodes as well.
492 */
493 switch (dip->di_aformat) {
494 case 0:
495 case XFS_DINODE_FMT_EXTENTS:
496 break;
497 default:
498 return __this_address;
499 }
500 if (dip->di_anextents)
501 return __this_address;
502 }
503
504 /* extent size hint validation */
505 fa = xfs_inode_validate_extsize(mp, be32_to_cpu(dip->di_extsize),
506 mode, flags);
507 if (fa)
508 return fa;
509
510 /* only version 3 or greater inodes are extensively verified here */
511 if (dip->di_version < 3)
512 return NULL;
513
514 flags2 = be64_to_cpu(dip->di_flags2);
515
516 /* don't allow reflink/cowextsize if we don't have reflink */
517 if ((flags2 & (XFS_DIFLAG2_REFLINK | XFS_DIFLAG2_COWEXTSIZE)) &&
518 !xfs_sb_version_hasreflink(&mp->m_sb))
519 return __this_address;
520
521 /* only regular files get reflink */
522 if ((flags2 & XFS_DIFLAG2_REFLINK) && (mode & S_IFMT) != S_IFREG)
523 return __this_address;
524
525 /* don't let reflink and realtime mix */
526 if ((flags2 & XFS_DIFLAG2_REFLINK) && (flags & XFS_DIFLAG_REALTIME))
527 return __this_address;
528
529 /* COW extent size hint validation */
530 fa = xfs_inode_validate_cowextsize(mp, be32_to_cpu(dip->di_cowextsize),
531 mode, flags, flags2);
532 if (fa)
533 return fa;
534
535 /* bigtime iflag can only happen on bigtime filesystems */
536 if (xfs_dinode_has_bigtime(dip) &&
537 !xfs_sb_version_hasbigtime(&mp->m_sb))
538 return __this_address;
539
540 return NULL;
541}
542
543void
544xfs_dinode_calc_crc(
545 struct xfs_mount *mp,
546 struct xfs_dinode *dip)
547{
548 uint32_t crc;
549
550 if (dip->di_version < 3)
551 return;
552
553 ASSERT(xfs_sb_version_hascrc(&mp->m_sb));
554 crc = xfs_start_cksum_update((char *)dip, mp->m_sb.sb_inodesize,
555 XFS_DINODE_CRC_OFF);
556 dip->di_crc = xfs_end_cksum(crc);
557}
558
559/*
560 * Validate di_extsize hint.
561 *
562 * 1. Extent size hint is only valid for directories and regular files.
563 * 2. FS_XFLAG_EXTSIZE is only valid for regular files.
564 * 3. FS_XFLAG_EXTSZINHERIT is only valid for directories.
565 * 4. Hint cannot be larger than MAXTEXTLEN.
566 * 5. Can be changed on directories at any time.
567 * 6. Hint value of 0 turns off hints, clears inode flags.
568 * 7. Extent size must be a multiple of the appropriate block size.
569 * For realtime files, this is the rt extent size.
570 * 8. For non-realtime files, the extent size hint must be limited
571 * to half the AG size to avoid alignment extending the extent beyond the
572 * limits of the AG.
573 */
574xfs_failaddr_t
575xfs_inode_validate_extsize(
576 struct xfs_mount *mp,
577 uint32_t extsize,
578 uint16_t mode,
579 uint16_t flags)
580{
581 bool rt_flag;
582 bool hint_flag;
583 bool inherit_flag;
584 uint32_t extsize_bytes;
585 uint32_t blocksize_bytes;
586
587 rt_flag = (flags & XFS_DIFLAG_REALTIME);
588 hint_flag = (flags & XFS_DIFLAG_EXTSIZE);
589 inherit_flag = (flags & XFS_DIFLAG_EXTSZINHERIT);
590 extsize_bytes = XFS_FSB_TO_B(mp, extsize);
591
592 /*
593 * This comment describes a historic gap in this verifier function.
594 *
595 * For a directory with both RTINHERIT and EXTSZINHERIT flags set, this
596 * function has never checked that the extent size hint is an integer
597 * multiple of the realtime extent size. Since we allow users to set
598 * this combination on non-rt filesystems /and/ to change the rt
599 * extent size when adding a rt device to a filesystem, the net effect
600 * is that users can configure a filesystem anticipating one rt
601 * geometry and change their minds later. Directories do not use the
602 * extent size hint, so this is harmless for them.
603 *
604 * If a directory with a misaligned extent size hint is allowed to
605 * propagate that hint into a new regular realtime file, the result
606 * is that the inode cluster buffer verifier will trigger a corruption
607 * shutdown the next time it is run, because the verifier has always
608 * enforced the alignment rule for regular files.
609 *
610 * Because we allow administrators to set a new rt extent size when
611 * adding a rt section, we cannot add a check to this verifier because
612 * that will result a new source of directory corruption errors when
613 * reading an existing filesystem. Instead, we rely on callers to
614 * decide when alignment checks are appropriate, and fix things up as
615 * needed.
616 */
617
618 if (rt_flag)
619 blocksize_bytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
620 else
621 blocksize_bytes = mp->m_sb.sb_blocksize;
622
623 if ((hint_flag || inherit_flag) && !(S_ISDIR(mode) || S_ISREG(mode)))
624 return __this_address;
625
626 if (hint_flag && !S_ISREG(mode))
627 return __this_address;
628
629 if (inherit_flag && !S_ISDIR(mode))
630 return __this_address;
631
632 if ((hint_flag || inherit_flag) && extsize == 0)
633 return __this_address;
634
635 /* free inodes get flags set to zero but extsize remains */
636 if (mode && !(hint_flag || inherit_flag) && extsize != 0)
637 return __this_address;
638
639 if (extsize_bytes % blocksize_bytes)
640 return __this_address;
641
642 if (extsize > MAXEXTLEN)
643 return __this_address;
644
645 if (!rt_flag && extsize > mp->m_sb.sb_agblocks / 2)
646 return __this_address;
647
648 return NULL;
649}
650
651/*
652 * Validate di_cowextsize hint.
653 *
654 * 1. CoW extent size hint can only be set if reflink is enabled on the fs.
655 * The inode does not have to have any shared blocks, but it must be a v3.
656 * 2. FS_XFLAG_COWEXTSIZE is only valid for directories and regular files;
657 * for a directory, the hint is propagated to new files.
658 * 3. Can be changed on files & directories at any time.
659 * 4. Hint value of 0 turns off hints, clears inode flags.
660 * 5. Extent size must be a multiple of the appropriate block size.
661 * 6. The extent size hint must be limited to half the AG size to avoid
662 * alignment extending the extent beyond the limits of the AG.
663 */
664xfs_failaddr_t
665xfs_inode_validate_cowextsize(
666 struct xfs_mount *mp,
667 uint32_t cowextsize,
668 uint16_t mode,
669 uint16_t flags,
670 uint64_t flags2)
671{
672 bool rt_flag;
673 bool hint_flag;
674 uint32_t cowextsize_bytes;
675
676 rt_flag = (flags & XFS_DIFLAG_REALTIME);
677 hint_flag = (flags2 & XFS_DIFLAG2_COWEXTSIZE);
678 cowextsize_bytes = XFS_FSB_TO_B(mp, cowextsize);
679
680 if (hint_flag && !xfs_sb_version_hasreflink(&mp->m_sb))
681 return __this_address;
682
683 if (hint_flag && !(S_ISDIR(mode) || S_ISREG(mode)))
684 return __this_address;
685
686 if (hint_flag && cowextsize == 0)
687 return __this_address;
688
689 /* free inodes get flags set to zero but cowextsize remains */
690 if (mode && !hint_flag && cowextsize != 0)
691 return __this_address;
692
693 if (hint_flag && rt_flag)
694 return __this_address;
695
696 if (cowextsize_bytes % mp->m_sb.sb_blocksize)
697 return __this_address;
698
699 if (cowextsize > MAXEXTLEN)
700 return __this_address;
701
702 if (cowextsize > mp->m_sb.sb_agblocks / 2)
703 return __this_address;
704
705 return NULL;
706}